How to check if the CUDA capable device is really working with MATLAB?

I have a program, where I want to check if is it possible to use CUDA calculation.

gpuDeviceCount

ans =

     1

says that I really do have a gpuDevice. However if I try to issue the

GPU_spectra=gpuArray(double(kiserletek.ROI_spektrumok(1).spectra));
There is a problem with the graphics driver or with this GPU device. Be
sure that you have a supported GPU and that the latest driver is
installed.

Caused by:
    The graphics driver was found, but it is too old. The graphics driver
    on your system supports CUDA version 6.5. The required CUDA version is
    9.1 or greater.
    
 

The command

    helyi_GPU=gpuDevice();
Error using gpuDevice (line 26)
There is a problem with the graphics driver or with this GPU device. Be
sure that you have a supported GPU and that the latest driver is
installed.

Caused by:
    The graphics driver was found, but it is too old. The graphics driver
    on your system supports CUDA version 6.5. The required CUDA version is
    9.1 or greater.
 

also gives an error.

It is OK, I understand it. This is my home computer where I have an old video card. At work I have a newer one.

BUT! How to check if my video card is capable of CUDA calculations in order to avoid the error message?????

Matlab version is: 2018b

[Walter Roberson]

What is the exact model of NVIDIA card you have?

[Csaba]

Geforce 210

It has 16 CUDA kernel, and there is no fresh driver. The latest is from 2016.12.14

Answer 1

[Walter Roberson]

The reference material indicates that the GeForce 210 was rather early and supported only cuda 1.2. MATLAB never supported 1.2. The first that MATLAB supported was 1.3 because it was 1.3 that added double precision support .

[Csaba]

Yes, thank you. I knew all you have written, and the question was not this.

The question is, that how can I avoid the error message, if, let say I want to check if there is a CUDA device available and do CUDA calculation if yes, and conventional calculation if not.

At present

gpuDeviceCount

gives me a non 0 answer for my old Geforce 210 but when the program wants utilize its presence runs to an error.

I use gpuDeviceCount to check if there is a capable CUDA device in the system or there is not, but it fails, so how can I do it other ways????

I have to add, that "before" my program was working all right, gpuDeviceCount gave me 0 if only GeForce 210 was present. I do not remember when this "before" was exactly, I am not sure that before I have upgraded to Matlab 2018b or not (I did not use this program for a while). I will reinstall Matlab 2017a (my previous release) and check it.

[Walter Roberson]

you might need try/catch

[Csaba]

So I have tried 2017b and - surprise - it was working as expected,

>> gpuDeviceCount

ans =

     0

This, I would say the "normal" behaviour. So it was changed in 2018b. I wonder if it is an intended change or just an accident?

I would prefer the 2017b version! GIVE ME BACK please!

[Csaba]

Dear Walter, I accept your answer (try/catch) because it really solves my problem, although I knew it before as well. BUT I have to add that this behaviour of gpuDeviceCount is an error IMHO.

[Walter Roberson]

I do not know if the change was deliberate . You can file a bug report .